Research And Application Of Map Matching And Synchronous Path Planning Method Based On High-precision Map

Map matching and path planning are one of the important research contents in the field of unmanned driving.However,due to the inherent shortcomings of traditional maps,the reliability of positioning and navigation will be affected.Therefore,the research on high-definition maps with higher accuracy and richer meaning is of profound significance.This paper studies map matching and path planning in the field of high-definition maps.Under the premise of fully understanding the characteristics and methods of point cloud data in high-definition maps,a novel U-Net network architecture based on conditional random fields is proposed.And then,the interference element segmentation and3D-NDT-based point cloud matching algorithm are combined to complete the matching positioning based on high-definition point cloud map.After that,this paper introduces the credibility evaluation of map matching into the generation of global guidance paths,which improves the security and reliability of path planning.The main work carried out in this article is as follows:1.A method based on conditional random field and improved U-Net for interference element segmentation is proposed and implemented.Map interference element segmentation is actually a point cloud semantic segmentation problem.First,a transformation method is used to transform a three-dimensional point cloud into a two-dimensional image by gridding in a polar coordinate system.The pooling parameters and activation functions of the U-Net network were modified.Then,according to the converted image characteristics,the pooling parameters and activation functions of the U-Net network were modified.For the problem that the original network could not output multi-classification results,the CRF model was introduced to process the classification results,and the semantic segmentation of the interfering elements in the point cloud map was completed.The network framework is concise and clear,and can adapt to the needs of real-time detection.Experiments on the KITTI data set prove that its performance is improved comparing to similar networks.2.A 3D-NDT point cloud map matching method based on map interference element filtering is proposed and implemented.Due to the interference of the scene in the high-definition map,it will affect the accuracy of map matching.In order to figure out this problem,this paper optimizes the process of filtering and then matching the interference elements in the point cloud,such as vehicles,pedestrians and cyclists.First,the point cloud is segmented using the improved U-Net network based on the conditional random field.The point cloud belonging to motor vehicles,pedestrians and cyclists is deleted from the current frame to obtain clear scene information.Then apply the 3D-NDT matching algorithm to locate the point cloud of the current frame into the global point cloud map to complete the map matching.Experiments on matching and positioning in multiple scenarios prove that this method can effectively improve the matching accuracy,in the meantime,avoid vehicle positioning errors caused by errors in the positioning system.3.An A~* path planning method which can make adaptive adjustment based on map matching credibility is proposed and implemented.When generating a guide path,most algorithms generally plan paths based on a known map,without considering the credibility of the map itself.This method introduces the correlation coefficient of the map matching effect into the cost function of A~* algorithm,so that the algorithm can generate a more secure insurance route in the case of different map credibility caused by different map matching effects.This method also introduces global steering information into the A~*algorithm heuristic function,which speeds up the algorithm’s convergence.Experiments in multiple scenarios prove that this method effectively improves the security and credibility of the generated guidance path.